net/sunrpc/xprtsock.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/net/sunrpc/xprtsock.c
   4  *
   5  * Client-side transport implementation for sockets.
   6  *
   7  * TCP callback races fixes (C) 1998 Red Hat
   8  * TCP send fixes (C) 1998 Red Hat
   9  * TCP NFS related read + write fixes
  10  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  11  *
  12  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  13  * Fix behaviour when socket buffer is full.
  14  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  15  *
  16  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  17  *
  18  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  19  *   <gilles.quillard@bull.net>
  20  */
  21
  22 #include <linux/types.h>
  23 #include <linux/string.h>
  24 #include <linux/slab.h>
  25 #include <linux/module.h>
  26 #include <linux/capability.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/errno.h>
  29 #include <linux/socket.h>
  30 #include <linux/in.h>
  31 #include <linux/net.h>
  32 #include <linux/mm.h>
  33 #include <linux/un.h>
  34 #include <linux/udp.h>
  35 #include <linux/tcp.h>
  36 #include <linux/sunrpc/clnt.h>
  37 #include <linux/sunrpc/addr.h>
  38 #include <linux/sunrpc/sched.h>
  39 #include <linux/sunrpc/svcsock.h>
  40 #include <linux/sunrpc/xprtsock.h>
  41 #include <linux/file.h>
  42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  43 #include <linux/sunrpc/bc_xprt.h>
  44 #endif
  45
  46 #include <net/sock.h>
  47 #include <net/checksum.h>
  48 #include <net/udp.h>
  49 #include <net/tcp.h>
  50
  51 #include <trace/events/sunrpc.h>
  52
  53 #include "sunrpc.h"
  54
  55 #define RPC_TCP_READ_CHUNK_SZ   (3*512*1024)
  56
  57 static void xs_close(struct rpc_xprt *xprt);
  58 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  59                 struct socket *sock);
  60
  61 /*
  62  * xprtsock tunables
  63  */
  64 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  65 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  66 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  67
  68 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  69 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  70
  71 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  72
  73 #define XS_TCP_LINGER_TO        (15U * HZ)
  74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  75
  76 /*
  77  * We can register our own files under /proc/sys/sunrpc by
  78  * calling register_sysctl_table() again.  The files in that
  79  * directory become the union of all files registered there.
  80  *
  81  * We simply need to make sure that we don't collide with
  82  * someone else's file names!
  83  */
  84
  85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  90
  91 static struct ctl_table_header *sunrpc_table_header;
  92
  93 /*
  94  * FIXME: changing the UDP slot table size should also resize the UDP
  95  *        socket buffers for existing UDP transports
  96  */
  97 static struct ctl_table xs_tunables_table[] = {
  98         {
  99                 .procname       = "udp_slot_table_entries",
 100                 .data           = &xprt_udp_slot_table_entries,
 101                 .maxlen         = sizeof(unsigned int),
 102                 .mode           = 0644,
 103                 .proc_handler   = proc_dointvec_minmax,
 104                 .extra1         = &min_slot_table_size,
 105                 .extra2         = &max_slot_table_size
 106         },
 107         {
 108                 .procname       = "tcp_slot_table_entries",
 109                 .data           = &xprt_tcp_slot_table_entries,
 110                 .maxlen         = sizeof(unsigned int),
 111                 .mode           = 0644,
 112                 .proc_handler   = proc_dointvec_minmax,
 113                 .extra1         = &min_slot_table_size,
 114                 .extra2         = &max_slot_table_size
 115         },
 116         {
 117                 .procname       = "tcp_max_slot_table_entries",
 118                 .data           = &xprt_max_tcp_slot_table_entries,
 119                 .maxlen         = sizeof(unsigned int),
 120                 .mode           = 0644,
 121                 .proc_handler   = proc_dointvec_minmax,
 122                 .extra1         = &min_slot_table_size,
 123                 .extra2         = &max_tcp_slot_table_limit
 124         },
 125         {
 126                 .procname       = "min_resvport",
 127                 .data           = &xprt_min_resvport,
 128                 .maxlen         = sizeof(unsigned int),
 129                 .mode           = 0644,
 130                 .proc_handler   = proc_dointvec_minmax,
 131                 .extra1         = &xprt_min_resvport_limit,
 132                 .extra2         = &xprt_max_resvport
 133         },
 134         {
 135                 .procname       = "max_resvport",
 136                 .data           = &xprt_max_resvport,
 137                 .maxlen         = sizeof(unsigned int),
 138                 .mode           = 0644,
 139                 .proc_handler   = proc_dointvec_minmax,
 140                 .extra1         = &xprt_min_resvport,
 141                 .extra2         = &xprt_max_resvport_limit
 142         },
 143         {
 144                 .procname       = "tcp_fin_timeout",
 145                 .data           = &xs_tcp_fin_timeout,
 146                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 147                 .mode           = 0644,
 148                 .proc_handler   = proc_dointvec_jiffies,
 149         },
 150         { },
 151 };
 152
 153 static struct ctl_table sunrpc_table[] = {
 154         {
 155                 .procname       = "sunrpc",
 156                 .mode           = 0555,
 157                 .child          = xs_tunables_table
 158         },
 159         { },
 160 };
 161
 162 #endif
 163
 164 /*
 165  * Wait duration for a reply from the RPC portmapper.
 166  */
 167 #define XS_BIND_TO              (60U * HZ)
 168
 169 /*
 170  * Delay if a UDP socket connect error occurs.  This is most likely some
 171  * kind of resource problem on the local host.
 172  */
 173 #define XS_UDP_REEST_TO         (2U * HZ)
 174
 175 /*
 176  * The reestablish timeout allows clients to delay for a bit before attempting
 177  * to reconnect to a server that just dropped our connection.
 178  *
 179  * We implement an exponential backoff when trying to reestablish a TCP
 180  * transport connection with the server.  Some servers like to drop a TCP
 181  * connection when they are overworked, so we start with a short timeout and
 182  * increase over time if the server is down or not responding.
 183  */
 184 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 185
 186 /*
 187  * TCP idle timeout; client drops the transport socket if it is idle
 188  * for this long.  Note that we also timeout UDP sockets to prevent
 189  * holding port numbers when there is no RPC traffic.
 190  */
 191 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 192
 193 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 194 # undef  RPC_DEBUG_DATA
 195 # define RPCDBG_FACILITY        RPCDBG_TRANS
 196 #endif
 197
 198 #ifdef RPC_DEBUG_DATA
 199 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 200 {
 201         u8 *buf = (u8 *) packet;
 202         int j;
 203
 204         dprintk("RPC:       %s\n", msg);
 205         for (j = 0; j < count && j < 128; j += 4) {
 206                 if (!(j & 31)) {
 207                         if (j)
 208                                 dprintk("\n");
 209                         dprintk("0x%04x ", j);
 210                 }
 211                 dprintk("%02x%02x%02x%02x ",
 212                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 213         }
 214         dprintk("\n");
 215 }
 216 #else
 217 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 218 {
 219         /* NOP */
 220 }
 221 #endif
 222
 223 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 224 {
 225         return (struct rpc_xprt *) sk->sk_user_data;
 226 }
 227
 228 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 229 {
 230         return (struct sockaddr *) &xprt->addr;
 231 }
 232
 233 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 234 {
 235         return (struct sockaddr_un *) &xprt->addr;
 236 }
 237
 238 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 239 {
 240         return (struct sockaddr_in *) &xprt->addr;
 241 }
 242
 243 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 244 {
 245         return (struct sockaddr_in6 *) &xprt->addr;
 246 }
 247
 248 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 249 {
 250         struct sockaddr *sap = xs_addr(xprt);
 251         struct sockaddr_in6 *sin6;
 252         struct sockaddr_in *sin;
 253         struct sockaddr_un *sun;
 254         char buf[128];
 255
 256         switch (sap->sa_family) {
 257         case AF_LOCAL:
 258                 sun = xs_addr_un(xprt);
 259                 strlcpy(buf, sun->sun_path, sizeof(buf));
 260                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 261                                                 kstrdup(buf, GFP_KERNEL);
 262                 break;
 263         case AF_INET:
 264                 (void)rpc_ntop(sap, buf, sizeof(buf));
 265                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 266                                                 kstrdup(buf, GFP_KERNEL);
 267                 sin = xs_addr_in(xprt);
 268                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 269                 break;
 270         case AF_INET6:
 271                 (void)rpc_ntop(sap, buf, sizeof(buf));
 272                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 273                                                 kstrdup(buf, GFP_KERNEL);
 274                 sin6 = xs_addr_in6(xprt);
 275                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 276                 break;
 277         default:
 278                 BUG();
 279         }
 280
 281         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 282 }
 283
 284 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 285 {
 286         struct sockaddr *sap = xs_addr(xprt);
 287         char buf[128];
 288
 289         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 290         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 291
 292         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 293         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 294 }
 295
 296 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 297                                      const char *protocol,
 298                                      const char *netid)
 299 {
 300         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 301         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 302         xs_format_common_peer_addresses(xprt);
 303         xs_format_common_peer_ports(xprt);
 304 }
 305
 306 static void xs_update_peer_port(struct rpc_xprt *xprt)
 307 {
 308         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 309         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 310
 311         xs_format_common_peer_ports(xprt);
 312 }
 313
 314 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 315 {
 316         unsigned int i;
 317
 318         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 319                 switch (i) {
 320                 case RPC_DISPLAY_PROTO:
 321                 case RPC_DISPLAY_NETID:
 322                         continue;
 323                 default:
 324                         kfree(xprt->address_strings[i]);
 325                 }
 326 }
 327
 328 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 329
 330 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 331 {
 332         struct msghdr msg = {
 333                 .msg_name       = addr,
 334                 .msg_namelen    = addrlen,
 335                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 336         };
 337         struct kvec iov = {
 338                 .iov_base       = vec->iov_base + base,
 339                 .iov_len        = vec->iov_len - base,
 340         };
 341
 342         if (iov.iov_len != 0)
 343                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 344         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 345 }
 346
 347 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 348 {
 349         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 350                         int offset, size_t size, int flags);
 351         struct page **ppage;
 352         unsigned int remainder;
 353         int err;
 354
 355         remainder = xdr->page_len - base;
 356         base += xdr->page_base;
 357         ppage = xdr->pages + (base >> PAGE_SHIFT);
 358         base &= ~PAGE_MASK;
 359         do_sendpage = sock->ops->sendpage;
 360         if (!zerocopy)
 361                 do_sendpage = sock_no_sendpage;
 362         for(;;) {
 363                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 364                 int flags = XS_SENDMSG_FLAGS;
 365
 366                 remainder -= len;
 367                 if (more)
 368                         flags |= MSG_MORE;
 369                 if (remainder != 0)
 370                         flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
 371                 err = do_sendpage(sock, *ppage, base, len, flags);
 372                 if (remainder == 0 || err != len)
 373                         break;
 374                 *sent_p += err;
 375                 ppage++;
 376                 base = 0;
 377         }
 378         if (err > 0) {
 379                 *sent_p += err;
 380                 err = 0;
 381         }
 382         return err;
 383 }
 384
 385 /**
 386  * xs_sendpages - write pages directly to a socket
 387  * @sock: socket to send on
 388  * @addr: UDP only -- address of destination
 389  * @addrlen: UDP only -- length of destination address
 390  * @xdr: buffer containing this request
 391  * @base: starting position in the buffer
 392  * @zerocopy: true if it is safe to use sendpage()
 393  * @sent_p: return the total number of bytes successfully queued for sending
 394  *
 395  */
 396 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 397 {
 398         unsigned int remainder = xdr->len - base;
 399         int err = 0;
 400         int sent = 0;
 401
 402         if (unlikely(!sock))
 403                 return -ENOTSOCK;
 404
 405         if (base != 0) {
 406                 addr = NULL;
 407                 addrlen = 0;
 408         }
 409
 410         if (base < xdr->head[0].iov_len || addr != NULL) {
 411                 unsigned int len = xdr->head[0].iov_len - base;
 412                 remainder -= len;
 413                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 414                 if (remainder == 0 || err != len)
 415                         goto out;
 416                 *sent_p += err;
 417                 base = 0;
 418         } else
 419                 base -= xdr->head[0].iov_len;
 420
 421         if (base < xdr->page_len) {
 422                 unsigned int len = xdr->page_len - base;
 423                 remainder -= len;
 424                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
 425                 *sent_p += sent;
 426                 if (remainder == 0 || sent != len)
 427                         goto out;
 428                 base = 0;
 429         } else
 430                 base -= xdr->page_len;
 431
 432         if (base >= xdr->tail[0].iov_len)
 433                 return 0;
 434         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 435 out:
 436         if (err > 0) {
 437                 *sent_p += err;
 438                 err = 0;
 439         }
 440         return err;
 441 }
 442
 443 static void xs_nospace_callback(struct rpc_task *task)
 444 {
 445         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 446
 447         transport->inet->sk_write_pending--;
 448 }
 449
 450 /**
 451  * xs_nospace - place task on wait queue if transmit was incomplete
 452  * @task: task to put to sleep
 453  *
 454  */
 455 static int xs_nospace(struct rpc_task *task)
 456 {
 457         struct rpc_rqst *req = task->tk_rqstp;
 458         struct rpc_xprt *xprt = req->rq_xprt;
 459         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 460         struct sock *sk = transport->inet;
 461         int ret = -EAGAIN;
 462
 463         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 464                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 465                         req->rq_slen);
 466
 467         /* Protect against races with write_space */
 468         spin_lock_bh(&xprt->transport_lock);
 469
 470         /* Don't race with disconnect */
 471         if (xprt_connected(xprt)) {
 472                 /* wait for more buffer space */
 473                 sk->sk_write_pending++;
 474                 xprt_wait_for_buffer_space(task, xs_nospace_callback);
 475         } else
 476                 ret = -ENOTCONN;
 477
 478         spin_unlock_bh(&xprt->transport_lock);
 479
 480         /* Race breaker in case memory is freed before above code is called */
 481         if (ret == -EAGAIN) {
 482                 struct socket_wq *wq;
 483
 484                 rcu_read_lock();
 485                 wq = rcu_dereference(sk->sk_wq);
 486                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
 487                 rcu_read_unlock();
 488
 489                 sk->sk_write_space(sk);
 490         }
 491         return ret;
 492 }
 493
 494 /*
 495  * Construct a stream transport record marker in @buf.
 496  */
 497 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 498 {
 499         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 500         rpc_fraghdr *base = buf->head[0].iov_base;
 501         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 502 }
 503
 504 /**
 505  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 506  * @task: RPC task that manages the state of an RPC request
 507  *
 508  * Return values:
 509  *        0:    The request has been sent
 510  *   EAGAIN:    The socket was blocked, please call again later to
 511  *              complete the request
 512  * ENOTCONN:    Caller needs to invoke connect logic then call again
 513  *    other:    Some other error occured, the request was not sent
 514  */
 515 static int xs_local_send_request(struct rpc_task *task)
 516 {
 517         struct rpc_rqst *req = task->tk_rqstp;
 518         struct rpc_xprt *xprt = req->rq_xprt;
 519         struct sock_xprt *transport =
 520                                 container_of(xprt, struct sock_xprt, xprt);
 521         struct xdr_buf *xdr = &req->rq_snd_buf;
 522         int status;
 523         int sent = 0;
 524
 525         xs_encode_stream_record_marker(&req->rq_snd_buf);
 526
 527         xs_pktdump("packet data:",
 528                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 529
 530         req->rq_xtime = ktime_get();
 531         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 532                               true, &sent);
 533         dprintk("RPC:       %s(%u) = %d\n",
 534                         __func__, xdr->len - req->rq_bytes_sent, status);
 535
 536         if (status == -EAGAIN && sock_writeable(transport->inet))
 537                 status = -ENOBUFS;
 538
 539         if (likely(sent > 0) || status == 0) {
 540                 req->rq_bytes_sent += sent;
 541                 req->rq_xmit_bytes_sent += sent;
 542                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 543                         req->rq_bytes_sent = 0;
 544                         return 0;
 545                 }
 546                 status = -EAGAIN;
 547         }
 548
 549         switch (status) {
 550         case -ENOBUFS:
 551                 break;
 552         case -EAGAIN:
 553                 status = xs_nospace(task);
 554                 break;
 555         default:
 556                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 557                         -status);
 558                 /* fall through */
 559         case -EPIPE:
 560                 xs_close(xprt);
 561                 status = -ENOTCONN;
 562         }
 563
 564         return status;
 565 }
 566
 567 /**
 568  * xs_udp_send_request - write an RPC request to a UDP socket
 569  * @task: address of RPC task that manages the state of an RPC request
 570  *
 571  * Return values:
 572  *        0:    The request has been sent
 573  *   EAGAIN:    The socket was blocked, please call again later to
 574  *              complete the request
 575  * ENOTCONN:    Caller needs to invoke connect logic then call again
 576  *    other:    Some other error occurred, the request was not sent
 577  */
 578 static int xs_udp_send_request(struct rpc_task *task)
 579 {
 580         struct rpc_rqst *req = task->tk_rqstp;
 581         struct rpc_xprt *xprt = req->rq_xprt;
 582         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 583         struct xdr_buf *xdr = &req->rq_snd_buf;
 584         int sent = 0;
 585         int status;
 586
 587         xs_pktdump("packet data:",
 588                                 req->rq_svec->iov_base,
 589                                 req->rq_svec->iov_len);
 590
 591         if (!xprt_bound(xprt))
 592                 return -ENOTCONN;
 593         req->rq_xtime = ktime_get();
 594         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 595                               xdr, req->rq_bytes_sent, true, &sent);
 596
 597         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 598                         xdr->len - req->rq_bytes_sent, status);
 599
 600         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 601         if (status == -EPERM)
 602                 goto process_status;
 603
 604         if (status == -EAGAIN && sock_writeable(transport->inet))
 605                 status = -ENOBUFS;
 606
 607         if (sent > 0 || status == 0) {
 608                 req->rq_xmit_bytes_sent += sent;
 609                 if (sent >= req->rq_slen)
 610                         return 0;
 611                 /* Still some bytes left; set up for a retry later. */
 612                 status = -EAGAIN;
 613         }
 614
 615 process_status:
 616         switch (status) {
 617         case -ENOTSOCK:
 618                 status = -ENOTCONN;
 619                 /* Should we call xs_close() here? */
 620                 break;
 621         case -EAGAIN:
 622                 status = xs_nospace(task);
 623                 break;
 624         case -ENETUNREACH:
 625         case -ENOBUFS:
 626         case -EPIPE:
 627         case -ECONNREFUSED:
 628         case -EPERM:
 629                 /* When the server has died, an ICMP port unreachable message
 630                  * prompts ECONNREFUSED. */
 631                 break;
 632         default:
 633                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 634                         -status);
 635         }
 636
 637         return status;
 638 }
 639
 640 /**
 641  * xs_tcp_send_request - write an RPC request to a TCP socket
 642  * @task: address of RPC task that manages the state of an RPC request
 643  *
 644  * Return values:
 645  *        0:    The request has been sent
 646  *   EAGAIN:    The socket was blocked, please call again later to
 647  *              complete the request
 648  * ENOTCONN:    Caller needs to invoke connect logic then call again
 649  *    other:    Some other error occurred, the request was not sent
 650  *
 651  * XXX: In the case of soft timeouts, should we eventually give up
 652  *      if sendmsg is not able to make progress?
 653  */
 654 static int xs_tcp_send_request(struct rpc_task *task)
 655 {
 656         struct rpc_rqst *req = task->tk_rqstp;
 657         struct rpc_xprt *xprt = req->rq_xprt;
 658         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 659         struct xdr_buf *xdr = &req->rq_snd_buf;
 660         bool zerocopy = true;
 661         bool vm_wait = false;
 662         int status;
 663         int sent;
 664
 665         xs_encode_stream_record_marker(&req->rq_snd_buf);
 666
 667         xs_pktdump("packet data:",
 668                                 req->rq_svec->iov_base,
 669                                 req->rq_svec->iov_len);
 670         /* Don't use zero copy if this is a resend. If the RPC call
 671          * completes while the socket holds a reference to the pages,
 672          * then we may end up resending corrupted data.
 673          */
 674         if (task->tk_flags & RPC_TASK_SENT)
 675                 zerocopy = false;
 676
 677         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
 678                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
 679
 680         /* Continue transmitting the packet/record. We must be careful
 681          * to cope with writespace callbacks arriving _after_ we have
 682          * called sendmsg(). */
 683         req->rq_xtime = ktime_get();
 684         while (1) {
 685                 sent = 0;
 686                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 687                                       req->rq_bytes_sent, zerocopy, &sent);
 688
 689                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 690                                 xdr->len - req->rq_bytes_sent, status);
 691
 692                 /* If we've sent the entire packet, immediately
 693                  * reset the count of bytes sent. */
 694                 req->rq_bytes_sent += sent;
 695                 req->rq_xmit_bytes_sent += sent;
 696                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 697                         req->rq_bytes_sent = 0;
 698                         return 0;
 699                 }
 700
 701                 WARN_ON_ONCE(sent == 0 && status == 0);
 702
 703                 if (status == -EAGAIN ) {
 704                         /*
 705                          * Return EAGAIN if we're sure we're hitting the
 706                          * socket send buffer limits.
 707                          */
 708                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 709                                 break;
 710                         /*
 711                          * Did we hit a memory allocation failure?
 712                          */
 713                         if (sent == 0) {
 714                                 status = -ENOBUFS;
 715                                 if (vm_wait)
 716                                         break;
 717                                 /* Retry, knowing now that we're below the
 718                                  * socket send buffer limit
 719                                  */
 720                                 vm_wait = true;
 721                         }
 722                         continue;
 723                 }
 724                 if (status < 0)
 725                         break;
 726                 vm_wait = false;
 727         }
 728
 729         switch (status) {
 730         case -ENOTSOCK:
 731                 status = -ENOTCONN;
 732                 /* Should we call xs_close() here? */
 733                 break;
 734         case -EAGAIN:
 735                 status = xs_nospace(task);
 736                 break;
 737         case -ECONNRESET:
 738         case -ECONNREFUSED:
 739         case -ENOTCONN:
 740         case -EADDRINUSE:
 741         case -ENOBUFS:
 742         case -EPIPE:
 743                 break;
 744         default:
 745                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 746                         -status);
 747         }
 748
 749         return status;
 750 }
 751
 752 /**
 753  * xs_tcp_release_xprt - clean up after a tcp transmission
 754  * @xprt: transport
 755  * @task: rpc task
 756  *
 757  * This cleans up if an error causes us to abort the transmission of a request.
 758  * In this case, the socket may need to be reset in order to avoid confusing
 759  * the server.
 760  */
 761 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 762 {
 763         struct rpc_rqst *req;
 764
 765         if (task != xprt->snd_task)
 766                 return;
 767         if (task == NULL)
 768                 goto out_release;
 769         req = task->tk_rqstp;
 770         if (req == NULL)
 771                 goto out_release;
 772         if (req->rq_bytes_sent == 0)
 773                 goto out_release;
 774         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 775                 goto out_release;
 776         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 777 out_release:
 778         xprt_release_xprt(xprt, task);
 779 }
 780
 781 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 782 {
 783         transport->old_data_ready = sk->sk_data_ready;
 784         transport->old_state_change = sk->sk_state_change;
 785         transport->old_write_space = sk->sk_write_space;
 786         transport->old_error_report = sk->sk_error_report;
 787 }
 788
 789 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 790 {
 791         sk->sk_data_ready = transport->old_data_ready;
 792         sk->sk_state_change = transport->old_state_change;
 793         sk->sk_write_space = transport->old_write_space;
 794         sk->sk_error_report = transport->old_error_report;
 795 }
 796
 797 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
 798 {
 799         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 800
 801         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 802 }
 803
 804 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 805 {
 806         smp_mb__before_atomic();
 807         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 808         clear_bit(XPRT_CLOSING, &xprt->state);
 809         xs_sock_reset_state_flags(xprt);
 810         smp_mb__after_atomic();
 811 }
 812
 813 /**
 814  * xs_error_report - callback to handle TCP socket state errors
 815  * @sk: socket
 816  *
 817  * Note: we don't call sock_error() since there may be a rpc_task
 818  * using the socket, and so we don't want to clear sk->sk_err.
 819  */
 820 static void xs_error_report(struct sock *sk)
 821 {
 822         struct rpc_xprt *xprt;
 823         int err;
 824
 825         read_lock_bh(&sk->sk_callback_lock);
 826         if (!(xprt = xprt_from_sock(sk)))
 827                 goto out;
 828
 829         err = -sk->sk_err;
 830         if (err == 0)
 831                 goto out;
 832         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 833                         xprt, -err);
 834         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 835         xprt_wake_pending_tasks(xprt, err);
 836  out:
 837         read_unlock_bh(&sk->sk_callback_lock);
 838 }
 839
 840 static void xs_reset_transport(struct sock_xprt *transport)
 841 {
 842         struct socket *sock = transport->sock;
 843         struct sock *sk = transport->inet;
 844         struct rpc_xprt *xprt = &transport->xprt;
 845
 846         if (sk == NULL)
 847                 return;
 848
 849         if (atomic_read(&transport->xprt.swapper))
 850                 sk_clear_memalloc(sk);
 851
 852         kernel_sock_shutdown(sock, SHUT_RDWR);
 853
 854         mutex_lock(&transport->recv_mutex);
 855         write_lock_bh(&sk->sk_callback_lock);
 856         transport->inet = NULL;
 857         transport->sock = NULL;
 858
 859         sk->sk_user_data = NULL;
 860
 861         xs_restore_old_callbacks(transport, sk);
 862         xprt_clear_connected(xprt);
 863         write_unlock_bh(&sk->sk_callback_lock);
 864         xs_sock_reset_connection_flags(xprt);
 865         mutex_unlock(&transport->recv_mutex);
 866
 867         trace_rpc_socket_close(xprt, sock);
 868         sock_release(sock);
 869 }
 870
 871 /**
 872  * xs_close - close a socket
 873  * @xprt: transport
 874  *
 875  * This is used when all requests are complete; ie, no DRC state remains
 876  * on the server we want to save.
 877  *
 878  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 879  * xs_reset_transport() zeroing the socket from underneath a writer.
 880  */
 881 static void xs_close(struct rpc_xprt *xprt)
 882 {
 883         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 884
 885         dprintk("RPC:       xs_close xprt %p\n", xprt);
 886
 887         xs_reset_transport(transport);
 888         xprt->reestablish_timeout = 0;
 889
 890         xprt_disconnect_done(xprt);
 891 }
 892
 893 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 894 {
 895         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 896                 xprt);
 897         xprt_disconnect_done(xprt);
 898 }
 899
 900 static void xs_xprt_free(struct rpc_xprt *xprt)
 901 {
 902         xs_free_peer_addresses(xprt);
 903         xprt_free(xprt);
 904 }
 905
 906 /**
 907  * xs_destroy - prepare to shutdown a transport
 908  * @xprt: doomed transport
 909  *
 910  */
 911 static void xs_destroy(struct rpc_xprt *xprt)
 912 {
 913         struct sock_xprt *transport = container_of(xprt,
 914                         struct sock_xprt, xprt);
 915         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 916
 917         cancel_delayed_work_sync(&transport->connect_worker);
 918         xs_close(xprt);
 919         cancel_work_sync(&transport->recv_worker);
 920         xs_xprt_free(xprt);
 921         module_put(THIS_MODULE);
 922 }
 923
 924 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 925 {
 926         struct xdr_skb_reader desc = {
 927                 .skb            = skb,
 928                 .offset         = sizeof(rpc_fraghdr),
 929                 .count          = skb->len - sizeof(rpc_fraghdr),
 930         };
 931
 932         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 933                 return -1;
 934         if (desc.count)
 935                 return -1;
 936         return 0;
 937 }
 938
 939 /**
 940  * xs_local_data_read_skb
 941  * @xprt: transport
 942  * @sk: socket
 943  * @skb: skbuff
 944  *
 945  * Currently this assumes we can read the whole reply in a single gulp.
 946  */
 947 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
 948                 struct sock *sk,
 949                 struct sk_buff *skb)
 950 {
 951         struct rpc_task *task;
 952         struct rpc_rqst *rovr;
 953         int repsize, copied;
 954         u32 _xid;
 955         __be32 *xp;
 956
 957         repsize = skb->len - sizeof(rpc_fraghdr);
 958         if (repsize < 4) {
 959                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 960                 return;
 961         }
 962
 963         /* Copy the XID from the skb... */
 964         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 965         if (xp == NULL)
 966                 return;
 967
 968         /* Look up and lock the request corresponding to the given XID */
 969         spin_lock(&xprt->recv_lock);
 970         rovr = xprt_lookup_rqst(xprt, *xp);
 971         if (!rovr)
 972                 goto out_unlock;
 973         xprt_pin_rqst(rovr);
 974         spin_unlock(&xprt->recv_lock);
 975         task = rovr->rq_task;
 976
 977         copied = rovr->rq_private_buf.buflen;
 978         if (copied > repsize)
 979                 copied = repsize;
 980
 981         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 982                 dprintk("RPC:       sk_buff copy failed\n");
 983                 spin_lock(&xprt->recv_lock);
 984                 goto out_unpin;
 985         }
 986
 987         spin_lock(&xprt->recv_lock);
 988         xprt_complete_rqst(task, copied);
 989 out_unpin:
 990         xprt_unpin_rqst(rovr);
 991  out_unlock:
 992         spin_unlock(&xprt->recv_lock);
 993 }
 994
 995 static void xs_local_data_receive(struct sock_xprt *transport)
 996 {
 997         struct sk_buff *skb;
 998         struct sock *sk;
 999         int err;
1000
1001 restart:
1002         mutex_lock(&transport->recv_mutex);
1003         sk = transport->inet;
1004         if (sk == NULL)
1005                 goto out;
1006         for (;;) {
1007                 skb = skb_recv_datagram(sk, 0, 1, &err);
1008                 if (skb != NULL) {
1009                         xs_local_data_read_skb(&transport->xprt, sk, skb);
1010                         skb_free_datagram(sk, skb);
1011                         continue;
1012                 }
1013                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1014                         break;
1015                 if (need_resched()) {
1016                         mutex_unlock(&transport->recv_mutex);
1017                         cond_resched();
1018                         goto restart;
1019                 }
1020         }
1021 out:
1022         mutex_unlock(&transport->recv_mutex);
1023 }
1024
1025 static void xs_local_data_receive_workfn(struct work_struct *work)
1026 {
1027         struct sock_xprt *transport =
1028                 container_of(work, struct sock_xprt, recv_worker);
1029         xs_local_data_receive(transport);
1030 }
1031
1032 /**
1033  * xs_udp_data_read_skb - receive callback for UDP sockets
1034  * @xprt: transport
1035  * @sk: socket
1036  * @skb: skbuff
1037  *
1038  */
1039 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1040                 struct sock *sk,
1041                 struct sk_buff *skb)
1042 {
1043         struct rpc_task *task;
1044         struct rpc_rqst *rovr;
1045         int repsize, copied;
1046         u32 _xid;
1047         __be32 *xp;
1048
1049         repsize = skb->len;
1050         if (repsize < 4) {
1051                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1052                 return;
1053         }
1054
1055         /* Copy the XID from the skb... */
1056         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1057         if (xp == NULL)
1058                 return;
1059
1060         /* Look up and lock the request corresponding to the given XID */
1061         spin_lock(&xprt->recv_lock);
1062         rovr = xprt_lookup_rqst(xprt, *xp);
1063         if (!rovr)
1064                 goto out_unlock;
1065         xprt_pin_rqst(rovr);
1066         xprt_update_rtt(rovr->rq_task);
1067         spin_unlock(&xprt->recv_lock);
1068         task = rovr->rq_task;
1069
1070         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1071                 copied = repsize;
1072
1073         /* Suck it into the iovec, verify checksum if not done by hw. */
1074         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1075                 spin_lock(&xprt->recv_lock);
1076                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1077                 goto out_unpin;
1078         }
1079
1080
1081         spin_lock_bh(&xprt->transport_lock);
1082         xprt_adjust_cwnd(xprt, task, copied);
1083         spin_unlock_bh(&xprt->transport_lock);
1084         spin_lock(&xprt->recv_lock);
1085         xprt_complete_rqst(task, copied);
1086         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1087 out_unpin:
1088         xprt_unpin_rqst(rovr);
1089  out_unlock:
1090         spin_unlock(&xprt->recv_lock);
1091 }
1092
1093 static void xs_udp_data_receive(struct sock_xprt *transport)
1094 {
1095         struct sk_buff *skb;
1096         struct sock *sk;
1097         int err;
1098
1099 restart:
1100         mutex_lock(&transport->recv_mutex);
1101         sk = transport->inet;
1102         if (sk == NULL)
1103                 goto out;
1104         for (;;) {
1105                 skb = skb_recv_udp(sk, 0, 1, &err);
1106                 if (skb != NULL) {
1107                         xs_udp_data_read_skb(&transport->xprt, sk, skb);
1108                         consume_skb(skb);
1109                         continue;
1110                 }
1111                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1112                         break;
1113                 if (need_resched()) {
1114                         mutex_unlock(&transport->recv_mutex);
1115                         cond_resched();
1116                         goto restart;
1117                 }
1118         }
1119 out:
1120         mutex_unlock(&transport->recv_mutex);
1121 }
1122
1123 static void xs_udp_data_receive_workfn(struct work_struct *work)
1124 {
1125         struct sock_xprt *transport =
1126                 container_of(work, struct sock_xprt, recv_worker);
1127         xs_udp_data_receive(transport);
1128 }
1129
1130 /**
1131  * xs_data_ready - "data ready" callback for UDP sockets
1132  * @sk: socket with data to read
1133  *
1134  */
1135 static void xs_data_ready(struct sock *sk)
1136 {
1137         struct rpc_xprt *xprt;
1138
1139         read_lock_bh(&sk->sk_callback_lock);
1140         dprintk("RPC:       xs_data_ready...\n");
1141         xprt = xprt_from_sock(sk);
1142         if (xprt != NULL) {
1143                 struct sock_xprt *transport = container_of(xprt,
1144                                 struct sock_xprt, xprt);
1145                 transport->old_data_ready(sk);
1146                 /* Any data means we had a useful conversation, so
1147                  * then we don't need to delay the next reconnect
1148                  */
1149                 if (xprt->reestablish_timeout)
1150                         xprt->reestablish_timeout = 0;
1151                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1152                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1153         }
1154         read_unlock_bh(&sk->sk_callback_lock);
1155 }
1156
1157 /*
1158  * Helper function to force a TCP close if the server is sending
1159  * junk and/or it has put us in CLOSE_WAIT
1160  */
1161 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1162 {
1163         xprt_force_disconnect(xprt);
1164 }
1165
1166 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1167 {
1168         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1169         size_t len, used;
1170         char *p;
1171
1172         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1173         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1174         used = xdr_skb_read_bits(desc, p, len);
1175         transport->tcp_offset += used;
1176         if (used != len)
1177                 return;
1178
1179         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1180         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1181                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1182         else
1183                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1184         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1185
1186         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1187         transport->tcp_offset = 0;
1188
1189         /* Sanity check of the record length */
1190         if (unlikely(transport->tcp_reclen < 8)) {
1191                 dprintk("RPC:       invalid TCP record fragment length\n");
1192                 xs_tcp_force_close(xprt);
1193                 return;
1194         }
1195         dprintk("RPC:       reading TCP record fragment of length %d\n",
1196                         transport->tcp_reclen);
1197 }
1198
1199 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1200 {
1201         if (transport->tcp_offset == transport->tcp_reclen) {
1202                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1203                 transport->tcp_offset = 0;
1204                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1205                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1206                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1207                         transport->tcp_copied = 0;
1208                 }
1209         }
1210 }
1211
1212 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1213 {
1214         size_t len, used;
1215         char *p;
1216
1217         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1218         dprintk("RPC:       reading XID (%zu bytes)\n", len);
1219         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1220         used = xdr_skb_read_bits(desc, p, len);
1221         transport->tcp_offset += used;
1222         if (used != len)
1223                 return;
1224         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1225         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1226         transport->tcp_copied = 4;
1227         dprintk("RPC:       reading %s XID %08x\n",
1228                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1229                                                               : "request with",
1230                         ntohl(transport->tcp_xid));
1231         xs_tcp_check_fraghdr(transport);
1232 }
1233
1234 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1235                                        struct xdr_skb_reader *desc)
1236 {
1237         size_t len, used;
1238         u32 offset;
1239         char *p;
1240
1241         /*
1242          * We want transport->tcp_offset to be 8 at the end of this routine
1243          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1244          * When this function is called for the first time,
1245          * transport->tcp_offset is 4 (after having already read the xid).
1246          */
1247         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1248         len = sizeof(transport->tcp_calldir) - offset;
1249         dprintk("RPC:       reading CALL/REPLY flag (%zu bytes)\n", len);
1250         p = ((char *) &transport->tcp_calldir) + offset;
1251         used = xdr_skb_read_bits(desc, p, len);
1252         transport->tcp_offset += used;
1253         if (used != len)
1254                 return;
1255         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1256         /*
1257          * We don't yet have the XDR buffer, so we will write the calldir
1258          * out after we get the buffer from the 'struct rpc_rqst'
1259          */
1260         switch (ntohl(transport->tcp_calldir)) {
1261         case RPC_REPLY:
1262                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1263                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1264                 transport->tcp_flags |= TCP_RPC_REPLY;
1265                 break;
1266         case RPC_CALL:
1267                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1268                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1269                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1270                 break;
1271         default:
1272                 dprintk("RPC:       invalid request message type\n");
1273                 xs_tcp_force_close(&transport->xprt);
1274         }
1275         xs_tcp_check_fraghdr(transport);
1276 }
1277
1278 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1279                                      struct xdr_skb_reader *desc,
1280                                      struct rpc_rqst *req)
1281 {
1282         struct sock_xprt *transport =
1283                                 container_of(xprt, struct sock_xprt, xprt);
1284         struct xdr_buf *rcvbuf;
1285         size_t len;
1286         ssize_t r;
1287
1288         rcvbuf = &req->rq_private_buf;
1289
1290         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1291                 /*
1292                  * Save the RPC direction in the XDR buffer
1293                  */
1294                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1295                         &transport->tcp_calldir,
1296                         sizeof(transport->tcp_calldir));
1297                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1298                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1299         }
1300
1301         len = desc->count;
1302         if (len > transport->tcp_reclen - transport->tcp_offset)
1303                 desc->count = transport->tcp_reclen - transport->tcp_offset;
1304         r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1305                                           desc, xdr_skb_read_bits);
1306
1307         if (desc->count) {
1308                 /* Error when copying to the receive buffer,
1309                  * usually because we weren't able to allocate
1310                  * additional buffer pages. All we can do now
1311                  * is turn off TCP_RCV_COPY_DATA, so the request
1312                  * will not receive any additional updates,
1313                  * and time out.
1314                  * Any remaining data from this record will
1315                  * be discarded.
1316                  */
1317                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1318                 dprintk("RPC:       XID %08x truncated request\n",
1319                                 ntohl(transport->tcp_xid));
1320                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1321                                 "tcp_offset = %u, tcp_reclen = %u\n",
1322                                 xprt, transport->tcp_copied,
1323                                 transport->tcp_offset, transport->tcp_reclen);
1324                 return;
1325         }
1326
1327         transport->tcp_copied += r;
1328         transport->tcp_offset += r;
1329         desc->count = len - r;
1330
1331         dprintk("RPC:       XID %08x read %zd bytes\n",
1332                         ntohl(transport->tcp_xid), r);
1333         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1334                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1335                         transport->tcp_offset, transport->tcp_reclen);
1336
1337         if (transport->tcp_copied == req->rq_private_buf.buflen)
1338                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1339         else if (transport->tcp_offset == transport->tcp_reclen) {
1340                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1341                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1342         }
1343 }
1344
1345 /*
1346  * Finds the request corresponding to the RPC xid and invokes the common
1347  * tcp read code to read the data.
1348  */
1349 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1350                                     struct xdr_skb_reader *desc)
1351 {
1352         struct sock_xprt *transport =
1353                                 container_of(xprt, struct sock_xprt, xprt);
1354         struct rpc_rqst *req;
1355
1356         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1357
1358         /* Find and lock the request corresponding to this xid */
1359         spin_lock(&xprt->recv_lock);
1360         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1361         if (!req) {
1362                 dprintk("RPC:       XID %08x request not found!\n",
1363                                 ntohl(transport->tcp_xid));
1364                 spin_unlock(&xprt->recv_lock);
1365                 return -1;
1366         }
1367         xprt_pin_rqst(req);
1368         spin_unlock(&xprt->recv_lock);
1369
1370         xs_tcp_read_common(xprt, desc, req);
1371
1372         spin_lock(&xprt->recv_lock);
1373         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1374                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1375         xprt_unpin_rqst(req);
1376         spin_unlock(&xprt->recv_lock);
1377         return 0;
1378 }
1379
1380 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1381 /*
1382  * Obtains an rpc_rqst previously allocated and invokes the common
1383  * tcp read code to read the data.  The result is placed in the callback
1384  * queue.
1385  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1386  * connection and return -1.
1387  */
1388 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1389                                        struct xdr_skb_reader *desc)
1390 {
1391         struct sock_xprt *transport =
1392                                 container_of(xprt, struct sock_xprt, xprt);
1393         struct rpc_rqst *req;
1394
1395         /* Look up the request corresponding to the given XID */
1396         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1397         if (req == NULL) {
1398                 printk(KERN_WARNING "Callback slot table overflowed\n");
1399                 xprt_force_disconnect(xprt);
1400                 return -1;
1401         }
1402
1403         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1404         xs_tcp_read_common(xprt, desc, req);
1405
1406         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1407                 xprt_complete_bc_request(req, transport->tcp_copied);
1408
1409         return 0;
1410 }
1411
1412 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1413                                         struct xdr_skb_reader *desc)
1414 {
1415         struct sock_xprt *transport =
1416                                 container_of(xprt, struct sock_xprt, xprt);
1417
1418         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1419                 xs_tcp_read_reply(xprt, desc) :
1420                 xs_tcp_read_callback(xprt, desc);
1421 }
1422
1423 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1424 {
1425         int ret;
1426
1427         ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1428                               SVC_SOCK_ANONYMOUS);
1429         if (ret < 0)
1430                 return ret;
1431         return 0;
1432 }
1433
1434 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1435 {
1436         return PAGE_SIZE;
1437 }
1438 #else
1439 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1440                                         struct xdr_skb_reader *desc)
1441 {
1442         return xs_tcp_read_reply(xprt, desc);
1443 }
1444 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1445
1446 /*
1447  * Read data off the transport.  This can be either an RPC_CALL or an
1448  * RPC_REPLY.  Relay the processing to helper functions.
1449  */
1450 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1451                                     struct xdr_skb_reader *desc)
1452 {
1453         struct sock_xprt *transport =
1454                                 container_of(xprt, struct sock_xprt, xprt);
1455
1456         if (_xs_tcp_read_data(xprt, desc) == 0)
1457                 xs_tcp_check_fraghdr(transport);
1458         else {
1459                 /*
1460                  * The transport_lock protects the request handling.
1461                  * There's no need to hold it to update the tcp_flags.
1462                  */
1463                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1464         }
1465 }
1466
1467 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1468 {
1469         size_t len;
1470
1471         len = transport->tcp_reclen - transport->tcp_offset;
1472         if (len > desc->count)
1473                 len = desc->count;
1474         desc->count -= len;
1475         desc->offset += len;
1476         transport->tcp_offset += len;
1477         dprintk("RPC:       discarded %zu bytes\n", len);
1478         xs_tcp_check_fraghdr(transport);
1479 }
1480
1481 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1482 {
1483         struct rpc_xprt *xprt = rd_desc->arg.data;
1484         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1485         struct xdr_skb_reader desc = {
1486                 .skb    = skb,
1487                 .offset = offset,
1488                 .count  = len,
1489         };
1490         size_t ret;
1491
1492         dprintk("RPC:       xs_tcp_data_recv started\n");
1493         do {
1494                 trace_xs_tcp_data_recv(transport);
1495                 /* Read in a new fragment marker if necessary */
1496                 /* Can we ever really expect to get completely empty fragments? */
1497                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1498                         xs_tcp_read_fraghdr(xprt, &desc);
1499                         continue;
1500                 }
1501                 /* Read in the xid if necessary */
1502                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1503                         xs_tcp_read_xid(transport, &desc);
1504                         continue;
1505                 }
1506                 /* Read in the call/reply flag */
1507                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1508                         xs_tcp_read_calldir(transport, &desc);
1509                         continue;
1510                 }
1511                 /* Read in the request data */
1512                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1513                         xs_tcp_read_data(xprt, &desc);
1514                         continue;
1515                 }
1516                 /* Skip over any trailing bytes on short reads */
1517                 xs_tcp_read_discard(transport, &desc);
1518         } while (desc.count);
1519         ret = len - desc.count;
1520         if (ret < rd_desc->count)
1521                 rd_desc->count -= ret;
1522         else
1523                 rd_desc->count = 0;
1524         trace_xs_tcp_data_recv(transport);
1525         dprintk("RPC:       xs_tcp_data_recv done\n");
1526         return ret;
1527 }
1528
1529 static void xs_tcp_data_receive(struct sock_xprt *transport)
1530 {
1531         struct rpc_xprt *xprt = &transport->xprt;
1532         struct sock *sk;
1533         read_descriptor_t rd_desc = {
1534                 .arg.data = xprt,
1535         };
1536         unsigned long total = 0;
1537         int read = 0;
1538
1539 restart:
1540         mutex_lock(&transport->recv_mutex);
1541         sk = transport->inet;
1542         if (sk == NULL)
1543                 goto out;
1544
1545         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1546         for (;;) {
1547                 rd_desc.count = RPC_TCP_READ_CHUNK_SZ;
1548                 lock_sock(sk);
1549                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1550                 if (rd_desc.count != 0 || read < 0) {
1551                         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1552                         release_sock(sk);
1553                         break;
1554                 }
1555                 release_sock(sk);
1556                 total += read;
1557                 if (need_resched()) {
1558                         mutex_unlock(&transport->recv_mutex);
1559                         cond_resched();
1560                         goto restart;
1561                 }
1562         }
1563         if (test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1564                 queue_work(xprtiod_workqueue, &transport->recv_worker);
1565 out:
1566         mutex_unlock(&transport->recv_mutex);
1567         trace_xs_tcp_data_ready(xprt, read, total);
1568 }
1569
1570 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1571 {
1572         struct sock_xprt *transport =
1573                 container_of(work, struct sock_xprt, recv_worker);
1574         xs_tcp_data_receive(transport);
1575 }
1576
1577 /**
1578  * xs_tcp_state_change - callback to handle TCP socket state changes
1579  * @sk: socket whose state has changed
1580  *
1581  */
1582 static void xs_tcp_state_change(struct sock *sk)
1583 {
1584         struct rpc_xprt *xprt;
1585         struct sock_xprt *transport;
1586
1587         read_lock_bh(&sk->sk_callback_lock);
1588         if (!(xprt = xprt_from_sock(sk)))
1589                 goto out;
1590         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1591         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1592                         sk->sk_state, xprt_connected(xprt),
1593                         sock_flag(sk, SOCK_DEAD),
1594                         sock_flag(sk, SOCK_ZAPPED),
1595                         sk->sk_shutdown);
1596
1597         transport = container_of(xprt, struct sock_xprt, xprt);
1598         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1599         switch (sk->sk_state) {
1600         case TCP_ESTABLISHED:
1601                 spin_lock(&xprt->transport_lock);
1602                 if (!xprt_test_and_set_connected(xprt)) {
1603
1604                         /* Reset TCP record info */
1605                         transport->tcp_offset = 0;
1606                         transport->tcp_reclen = 0;
1607                         transport->tcp_copied = 0;
1608                         transport->tcp_flags =
1609                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1610                         xprt->connect_cookie++;
1611                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1612                         xprt_clear_connecting(xprt);
1613
1614                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1615                 }
1616                 spin_unlock(&xprt->transport_lock);
1617                 break;
1618         case TCP_FIN_WAIT1:
1619                 /* The client initiated a shutdown of the socket */
1620                 xprt->connect_cookie++;
1621                 xprt->reestablish_timeout = 0;
1622                 set_bit(XPRT_CLOSING, &xprt->state);
1623                 smp_mb__before_atomic();
1624                 clear_bit(XPRT_CONNECTED, &xprt->state);
1625                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1626                 smp_mb__after_atomic();
1627                 break;
1628         case TCP_CLOSE_WAIT:
1629                 /* The server initiated a shutdown of the socket */
1630                 xprt->connect_cookie++;
1631                 clear_bit(XPRT_CONNECTED, &xprt->state);
1632                 xs_tcp_force_close(xprt);
1633                 /* fall through */
1634         case TCP_CLOSING:
1635                 /*
1636                  * If the server closed down the connection, make sure that
1637                  * we back off before reconnecting
1638                  */
1639                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1640                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1641                 break;
1642         case TCP_LAST_ACK:
1643                 set_bit(XPRT_CLOSING, &xprt->state);
1644                 smp_mb__before_atomic();
1645                 clear_bit(XPRT_CONNECTED, &xprt->state);
1646                 smp_mb__after_atomic();
1647                 break;
1648         case TCP_CLOSE:
1649                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1650                                         &transport->sock_state))
1651                         xprt_clear_connecting(xprt);
1652                 clear_bit(XPRT_CLOSING, &xprt->state);
1653                 if (sk->sk_err)
1654                         xprt_wake_pending_tasks(xprt, -sk->sk_err);
1655                 /* Trigger the socket release */
1656                 xs_tcp_force_close(xprt);
1657         }
1658  out:
1659         read_unlock_bh(&sk->sk_callback_lock);
1660 }
1661
1662 static void xs_write_space(struct sock *sk)
1663 {
1664         struct socket_wq *wq;
1665         struct rpc_xprt *xprt;
1666
1667         if (!sk->sk_socket)
1668                 return;
1669         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1670
1671         if (unlikely(!(xprt = xprt_from_sock(sk))))
1672                 return;
1673         rcu_read_lock();
1674         wq = rcu_dereference(sk->sk_wq);
1675         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1676                 goto out;
1677
1678         xprt_write_space(xprt);
1679 out:
1680         rcu_read_unlock();
1681 }
1682
1683 /**
1684  * xs_udp_write_space - callback invoked when socket buffer space
1685  *                             becomes available
1686  * @sk: socket whose state has changed
1687  *
1688  * Called when more output buffer space is available for this socket.
1689  * We try not to wake our writers until they can make "significant"
1690  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1691  * with a bunch of small requests.
1692  */
1693 static void xs_udp_write_space(struct sock *sk)
1694 {
1695         read_lock_bh(&sk->sk_callback_lock);
1696
1697         /* from net/core/sock.c:sock_def_write_space */
1698         if (sock_writeable(sk))
1699                 xs_write_space(sk);
1700
1701         read_unlock_bh(&sk->sk_callback_lock);
1702 }
1703
1704 /**
1705  * xs_tcp_write_space - callback invoked when socket buffer space
1706  *                             becomes available
1707  * @sk: socket whose state has changed
1708  *
1709  * Called when more output buffer space is available for this socket.
1710  * We try not to wake our writers until they can make "significant"
1711  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1712  * with a bunch of small requests.
1713  */
1714 static void xs_tcp_write_space(struct sock *sk)
1715 {
1716         read_lock_bh(&sk->sk_callback_lock);
1717
1718         /* from net/core/stream.c:sk_stream_write_space */
1719         if (sk_stream_is_writeable(sk))
1720                 xs_write_space(sk);
1721
1722         read_unlock_bh(&sk->sk_callback_lock);
1723 }
1724
1725 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1726 {
1727         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1728         struct sock *sk = transport->inet;
1729
1730         if (transport->rcvsize) {
1731                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1732                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1733         }
1734         if (transport->sndsize) {
1735                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1736                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1737                 sk->sk_write_space(sk);
1738         }
1739 }
1740
1741 /**
1742  * xs_udp_set_buffer_size - set send and receive limits
1743  * @xprt: generic transport
1744  * @sndsize: requested size of send buffer, in bytes
1745  * @rcvsize: requested size of receive buffer, in bytes
1746  *
1747  * Set socket send and receive buffer size limits.
1748  */
1749 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1750 {
1751         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1752
1753         transport->sndsize = 0;
1754         if (sndsize)
1755                 transport->sndsize = sndsize + 1024;
1756         transport->rcvsize = 0;
1757         if (rcvsize)
1758                 transport->rcvsize = rcvsize + 1024;
1759
1760         xs_udp_do_set_buffer_size(xprt);
1761 }
1762
1763 /**
1764  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1765  * @task: task that timed out
1766  *
1767  * Adjust the congestion window after a retransmit timeout has occurred.
1768  */
1769 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1770 {
1771         spin_lock_bh(&xprt->transport_lock);
1772         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1773         spin_unlock_bh(&xprt->transport_lock);
1774 }
1775
1776 static unsigned short xs_get_random_port(void)
1777 {
1778         unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
1779         unsigned short rand = (unsigned short) prandom_u32() % range;
1780         return rand + xprt_min_resvport;
1781 }
1782
1783 /**
1784  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1785  * @sock: socket
1786  *
1787  * Note that this function has to be called on all sockets that share the
1788  * same port, and it must be called before binding.
1789  */
1790 static void xs_sock_set_reuseport(struct socket *sock)
1791 {
1792         int opt = 1;
1793
1794         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1795                         (char *)&opt, sizeof(opt));
1796 }
1797
1798 static unsigned short xs_sock_getport(struct socket *sock)
1799 {
1800         struct sockaddr_storage buf;
1801         unsigned short port = 0;
1802
1803         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1804                 goto out;
1805         switch (buf.ss_family) {
1806         case AF_INET6:
1807                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1808                 break;
1809         case AF_INET:
1810                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1811         }
1812 out:
1813         return port;
1814 }
1815
1816 /**
1817  * xs_set_port - reset the port number in the remote endpoint address
1818  * @xprt: generic transport
1819  * @port: new port number
1820  *
1821  */
1822 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1823 {
1824         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1825
1826         rpc_set_port(xs_addr(xprt), port);
1827         xs_update_peer_port(xprt);
1828 }
1829
1830 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1831 {
1832         if (transport->srcport == 0)
1833                 transport->srcport = xs_sock_getport(sock);
1834 }
1835
1836 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1837 {
1838         unsigned short port = transport->srcport;
1839
1840         if (port == 0 && transport->xprt.resvport)
1841                 port = xs_get_random_port();
1842         return port;
1843 }
1844
1845 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1846 {
1847         if (transport->srcport != 0)
1848                 transport->srcport = 0;
1849         if (!transport->xprt.resvport)
1850                 return 0;
1851         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1852                 return xprt_max_resvport;
1853         return --port;
1854 }
1855 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1856 {
1857         struct sockaddr_storage myaddr;
1858         int err, nloop = 0;
1859         unsigned short port = xs_get_srcport(transport);
1860         unsigned short last;
1861
1862         /*
1863          * If we are asking for any ephemeral port (i.e. port == 0 &&
1864          * transport->xprt.resvport == 0), don't bind.  Let the local
1865          * port selection happen implicitly when the socket is used
1866          * (for example at connect time).
1867          *
1868          * This ensures that we can continue to establish TCP
1869          * connections even when all local ephemeral ports are already
1870          * a part of some TCP connection.  This makes no difference
1871          * for UDP sockets, but also doens't harm them.
1872          *
1873          * If we're asking for any reserved port (i.e. port == 0 &&
1874          * transport->xprt.resvport == 1) xs_get_srcport above will
1875          * ensure that port is non-zero and we will bind as needed.
1876          */
1877         if (port == 0)
1878                 return 0;
1879
1880         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1881         do {
1882                 rpc_set_port((struct sockaddr *)&myaddr, port);
1883                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1884                                 transport->xprt.addrlen);
1885                 if (err == 0) {
1886                         transport->srcport = port;
1887                         break;
1888                 }
1889                 last = port;
1890                 port = xs_next_srcport(transport, port);
1891                 if (port > last)
1892                         nloop++;
1893         } while (err == -EADDRINUSE && nloop != 2);
1894
1895         if (myaddr.ss_family == AF_INET)
1896                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1897                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1898                                 port, err ? "failed" : "ok", err);
1899         else
1900                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1901                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1902                                 port, err ? "failed" : "ok", err);
1903         return err;
1904 }
1905
1906 /*
1907  * We don't support autobind on AF_LOCAL sockets
1908  */
1909 static void xs_local_rpcbind(struct rpc_task *task)
1910 {
1911         xprt_set_bound(task->tk_xprt);
1912 }
1913
1914 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1915 {
1916 }
1917
1918 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1919 static struct lock_class_key xs_key[2];
1920 static struct lock_class_key xs_slock_key[2];
1921
1922 static inline void xs_reclassify_socketu(struct socket *sock)
1923 {
1924         struct sock *sk = sock->sk;
1925
1926         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1927                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1928 }
1929
1930 static inline void xs_reclassify_socket4(struct socket *sock)
1931 {
1932         struct sock *sk = sock->sk;
1933
1934         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1935                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1936 }
1937
1938 static inline void xs_reclassify_socket6(struct socket *sock)
1939 {
1940         struct sock *sk = sock->sk;
1941
1942         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1943                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1944 }
1945
1946 static inline void xs_reclassify_socket(int family, struct socket *sock)
1947 {
1948         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1949                 return;
1950
1951         switch (family) {
1952         case AF_LOCAL:
1953                 xs_reclassify_socketu(sock);
1954                 break;
1955         case AF_INET:
1956                 xs_reclassify_socket4(sock);
1957                 break;
1958         case AF_INET6:
1959                 xs_reclassify_socket6(sock);
1960                 break;
1961         }
1962 }
1963 #else
1964 static inline void xs_reclassify_socket(int family, struct socket *sock)
1965 {
1966 }
1967 #endif
1968
1969 static void xs_dummy_setup_socket(struct work_struct *work)
1970 {
1971 }
1972
1973 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1974                 struct sock_xprt *transport, int family, int type,
1975                 int protocol, bool reuseport)
1976 {
1977         struct socket *sock;
1978         int err;
1979
1980         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1981         if (err < 0) {
1982                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1983                                 protocol, -err);
1984                 goto out;
1985         }
1986         xs_reclassify_socket(family, sock);
1987
1988         if (reuseport)
1989                 xs_sock_set_reuseport(sock);
1990
1991         err = xs_bind(transport, sock);
1992         if (err) {
1993                 sock_release(sock);
1994                 goto out;
1995         }
1996
1997         return sock;
1998 out:
1999         return ERR_PTR(err);
2000 }
2001
2002 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
2003                                       struct socket *sock)
2004 {
2005         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2006                                                                         xprt);
2007
2008         if (!transport->inet) {
2009                 struct sock *sk = sock->sk;
2010
2011                 write_lock_bh(&sk->sk_callback_lock);
2012
2013                 xs_save_old_callbacks(transport, sk);
2014
2015                 sk->sk_user_data = xprt;
2016                 sk->sk_data_ready = xs_data_ready;
2017                 sk->sk_write_space = xs_udp_write_space;
2018                 sock_set_flag(sk, SOCK_FASYNC);
2019                 sk->sk_error_report = xs_error_report;
2020                 sk->sk_allocation = GFP_NOIO;
2021
2022                 xprt_clear_connected(xprt);
2023
2024                 /* Reset to new socket */
2025                 transport->sock = sock;
2026                 transport->inet = sk;
2027
2028                 write_unlock_bh(&sk->sk_callback_lock);
2029         }
2030
2031         /* Tell the socket layer to start connecting... */
2032         xprt->stat.connect_count++;
2033         xprt->stat.connect_start = jiffies;
2034         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2035 }
2036
2037 /**
2038  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2039  * @transport: socket transport to connect
2040  */
2041 static int xs_local_setup_socket(struct sock_xprt *transport)
2042 {
2043         struct rpc_xprt *xprt = &transport->xprt;
2044         struct socket *sock;
2045         int status = -EIO;
2046
2047         status = __sock_create(xprt->xprt_net, AF_LOCAL,
2048                                         SOCK_STREAM, 0, &sock, 1);
2049         if (status < 0) {
2050                 dprintk("RPC:       can't create AF_LOCAL "
2051                         "transport socket (%d).\n", -status);
2052                 goto out;
2053         }
2054         xs_reclassify_socket(AF_LOCAL, sock);
2055
2056         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2057                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2058
2059         status = xs_local_finish_connecting(xprt, sock);
2060         trace_rpc_socket_connect(xprt, sock, status);
2061         switch (status) {
2062         case 0:
2063                 dprintk("RPC:       xprt %p connected to %s\n",
2064                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2065                 xprt_set_connected(xprt);
2066         case -ENOBUFS:
2067                 break;
2068         case -ENOENT:
2069                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2070                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2071                 break;
2072         case -ECONNREFUSED:
2073                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2074                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2075                 break;
2076         default:
2077                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2078                                 __func__, -status,
2079                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2080         }
2081
2082 out:
2083         xprt_clear_connecting(xprt);
2084         xprt_wake_pending_tasks(xprt, status);
2085         return status;
2086 }
2087
2088 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2089 {
2090         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2091         int ret;
2092
2093          if (RPC_IS_ASYNC(task)) {
2094                 /*
2095                  * We want the AF_LOCAL connect to be resolved in the
2096                  * filesystem namespace of the process making the rpc
2097                  * call.  Thus we connect synchronously.
2098                  *
2099                  * If we want to support asynchronous AF_LOCAL calls,
2100                  * we'll need to figure out how to pass a namespace to
2101                  * connect.
2102                  */
2103                 rpc_exit(task, -ENOTCONN);
2104                 return;
2105         }
2106         ret = xs_local_setup_socket(transport);
2107         if (ret && !RPC_IS_SOFTCONN(task))
2108                 msleep_interruptible(15000);
2109 }
2110
2111 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2112 /*
2113  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2114  * know that we have exclusive access to the socket), to guard against
2115  * races with xs_reset_transport.
2116  */
2117 static void xs_set_memalloc(struct rpc_xprt *xprt)
2118 {
2119         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2120                         xprt);
2121
2122         /*
2123          * If there's no sock, then we have nothing to set. The
2124          * reconnecting process will get it for us.
2125          */
2126         if (!transport->inet)
2127                 return;
2128         if (atomic_read(&xprt->swapper))
2129                 sk_set_memalloc(transport->inet);
2130 }
2131
2132 /**
2133  * xs_enable_swap - Tag this transport as being used for swap.
2134  * @xprt: transport to tag
2135  *
2136  * Take a reference to this transport on behalf of the rpc_clnt, and
2137  * optionally mark it for swapping if it wasn't already.
2138  */
2139 static int
2140 xs_enable_swap(struct rpc_xprt *xprt)
2141 {
2142         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2143
2144         if (atomic_inc_return(&xprt->swapper) != 1)
2145                 return 0;
2146         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2147                 return -ERESTARTSYS;
2148         if (xs->inet)
2149                 sk_set_memalloc(xs->inet);
2150         xprt_release_xprt(xprt, NULL);
2151         return 0;
2152 }
2153
2154 /**
2155  * xs_disable_swap - Untag this transport as being used for swap.
2156  * @xprt: transport to tag
2157  *
2158  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2159  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2160  */
2161 static void
2162 xs_disable_swap(struct rpc_xprt *xprt)
2163 {
2164         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2165
2166         if (!atomic_dec_and_test(&xprt->swapper))
2167                 return;
2168         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2169                 return;
2170         if (xs->inet)
2171                 sk_clear_memalloc(xs->inet);
2172         xprt_release_xprt(xprt, NULL);
2173 }
2174 #else
2175 static void xs_set_memalloc(struct rpc_xprt *xprt)
2176 {
2177 }
2178
2179 static int
2180 xs_enable_swap(struct rpc_xprt *xprt)
2181 {
2182         return -EINVAL;
2183 }
2184
2185 static void
2186 xs_disable_swap(struct rpc_xprt *xprt)
2187 {
2188 }
2189 #endif
2190
2191 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2192 {
2193         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2194
2195         if (!transport->inet) {
2196                 struct sock *sk = sock->sk;
2197
2198                 write_lock_bh(&sk->sk_callback_lock);
2199
2200                 xs_save_old_callbacks(transport, sk);
2201
2202                 sk->sk_user_data = xprt;
2203                 sk->sk_data_ready = xs_data_ready;
2204                 sk->sk_write_space = xs_udp_write_space;
2205                 sock_set_flag(sk, SOCK_FASYNC);
2206                 sk->sk_allocation = GFP_NOIO;
2207
2208                 xprt_set_connected(xprt);
2209
2210                 /* Reset to new socket */
2211                 transport->sock = sock;
2212                 transport->inet = sk;
2213
2214                 xs_set_memalloc(xprt);
2215
2216                 write_unlock_bh(&sk->sk_callback_lock);
2217         }
2218         xs_udp_do_set_buffer_size(xprt);
2219
2220         xprt->stat.connect_start = jiffies;
2221 }
2222
2223 static void xs_udp_setup_socket(struct work_struct *work)
2224 {
2225         struct sock_xprt *transport =
2226                 container_of(work, struct sock_xprt, connect_worker.work);
2227         struct rpc_xprt *xprt = &transport->xprt;
2228         struct socket *sock;
2229         int status = -EIO;
2230
2231         sock = xs_create_sock(xprt, transport,
2232                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2233                         IPPROTO_UDP, false);
2234         if (IS_ERR(sock))
2235                 goto out;
2236
2237         dprintk("RPC:       worker connecting xprt %p via %s to "
2238                                 "%s (port %s)\n", xprt,
2239                         xprt->address_strings[RPC_DISPLAY_PROTO],
2240                         xprt->address_strings[RPC_DISPLAY_ADDR],
2241                         xprt->address_strings[RPC_DISPLAY_PORT]);
2242
2243         xs_udp_finish_connecting(xprt, sock);
2244         trace_rpc_socket_connect(xprt, sock, 0);
2245         status = 0;
2246 out:
2247         xprt_unlock_connect(xprt, transport);
2248         xprt_clear_connecting(xprt);
2249         xprt_wake_pending_tasks(xprt, status);
2250 }
2251
2252 /**
2253  * xs_tcp_shutdown - gracefully shut down a TCP socket
2254  * @xprt: transport
2255  *
2256  * Initiates a graceful shutdown of the TCP socket by calling the
2257  * equivalent of shutdown(SHUT_RDWR);
2258  */
2259 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2260 {
2261         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2262         struct socket *sock = transport->sock;
2263         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2264
2265         if (sock == NULL)
2266                 return;
2267         switch (skst) {
2268         default:
2269                 kernel_sock_shutdown(sock, SHUT_RDWR);
2270                 trace_rpc_socket_shutdown(xprt, sock);
2271                 break;
2272         case TCP_CLOSE:
2273         case TCP_TIME_WAIT:
2274                 xs_reset_transport(transport);
2275         }
2276 }
2277
2278 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2279                 struct socket *sock)
2280 {
2281         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2282         unsigned int keepidle;
2283         unsigned int keepcnt;
2284         unsigned int opt_on = 1;
2285         unsigned int timeo;
2286
2287         spin_lock_bh(&xprt->transport_lock);
2288         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2289         keepcnt = xprt->timeout->to_retries + 1;
2290         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2291                 (xprt->timeout->to_retries + 1);
2292         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2293         spin_unlock_bh(&xprt->transport_lock);
2294
2295         /* TCP Keepalive options */
2296         kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2297                         (char *)&opt_on, sizeof(opt_on));
2298         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2299                         (char *)&keepidle, sizeof(keepidle));
2300         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2301                         (char *)&keepidle, sizeof(keepidle));
2302         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2303                         (char *)&keepcnt, sizeof(keepcnt));
2304
2305         /* TCP user timeout (see RFC5482) */
2306         kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2307                         (char *)&timeo, sizeof(timeo));
2308 }
2309
2310 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2311                 unsigned long connect_timeout,
2312                 unsigned long reconnect_timeout)
2313 {
2314         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2315         struct rpc_timeout to;
2316         unsigned long initval;
2317
2318         spin_lock_bh(&xprt->transport_lock);
2319         if (reconnect_timeout < xprt->max_reconnect_timeout)
2320                 xprt->max_reconnect_timeout = reconnect_timeout;
2321         if (connect_timeout < xprt->connect_timeout) {
2322                 memcpy(&to, xprt->timeout, sizeof(to));
2323                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2324                 /* Arbitrary lower limit */
2325                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2326                         initval = XS_TCP_INIT_REEST_TO << 1;
2327                 to.to_initval = initval;
2328                 to.to_maxval = initval;
2329                 memcpy(&transport->tcp_timeout, &to,
2330                                 sizeof(transport->tcp_timeout));
2331                 xprt->timeout = &transport->tcp_timeout;
2332                 xprt->connect_timeout = connect_timeout;
2333         }
2334         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2335         spin_unlock_bh(&xprt->transport_lock);
2336 }
2337
2338 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2339 {
2340         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2341         int ret = -ENOTCONN;
2342
2343         if (!transport->inet) {
2344                 struct sock *sk = sock->sk;
2345                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2346
2347                 /* Avoid temporary address, they are bad for long-lived
2348                  * connections such as NFS mounts.
2349                  * RFC4941, section 3.6 suggests that:
2350                  *    Individual applications, which have specific
2351                  *    knowledge about the normal duration of connections,
2352                  *    MAY override this as appropriate.
2353                  */
2354                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2355                                 (char *)&addr_pref, sizeof(addr_pref));
2356
2357                 xs_tcp_set_socket_timeouts(xprt, sock);
2358
2359                 write_lock_bh(&sk->sk_callback_lock);
2360
2361                 xs_save_old_callbacks(transport, sk);
2362
2363                 sk->sk_user_data = xprt;
2364                 sk->sk_data_ready = xs_data_ready;
2365                 sk->sk_state_change = xs_tcp_state_change;
2366                 sk->sk_write_space = xs_tcp_write_space;
2367                 sock_set_flag(sk, SOCK_FASYNC);
2368                 sk->sk_error_report = xs_error_report;
2369                 sk->sk_allocation = GFP_NOIO;
2370
2371                 /* socket options */
2372                 sock_reset_flag(sk, SOCK_LINGER);
2373                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2374
2375                 xprt_clear_connected(xprt);
2376
2377                 /* Reset to new socket */
2378                 transport->sock = sock;
2379                 transport->inet = sk;
2380
2381                 write_unlock_bh(&sk->sk_callback_lock);
2382         }
2383
2384         if (!xprt_bound(xprt))
2385                 goto out;
2386
2387         xs_set_memalloc(xprt);
2388
2389         /* Tell the socket layer to start connecting... */
2390         xprt->stat.connect_count++;
2391         xprt->stat.connect_start = jiffies;
2392         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2393         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2394         switch (ret) {
2395         case 0:
2396                 xs_set_srcport(transport, sock);
2397                 /* fall through */
2398         case -EINPROGRESS:
2399                 /* SYN_SENT! */
2400                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2401                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2402                 break;
2403         case -EADDRNOTAVAIL:
2404                 /* Source port number is unavailable. Try a new one! */
2405                 transport->srcport = 0;
2406         }
2407 out:
2408         return ret;
2409 }
2410
2411 /**
2412  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2413  *
2414  * Invoked by a work queue tasklet.
2415  */
2416 static void xs_tcp_setup_socket(struct work_struct *work)
2417 {
2418         struct sock_xprt *transport =
2419                 container_of(work, struct sock_xprt, connect_worker.work);
2420         struct socket *sock = transport->sock;
2421         struct rpc_xprt *xprt = &transport->xprt;
2422         int status = -EIO;
2423
2424         if (!sock) {
2425                 sock = xs_create_sock(xprt, transport,
2426                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2427                                 IPPROTO_TCP, true);
2428                 if (IS_ERR(sock)) {
2429                         status = PTR_ERR(sock);
2430                         goto out;
2431                 }
2432         }
2433
2434         dprintk("RPC:       worker connecting xprt %p via %s to "
2435                                 "%s (port %s)\n", xprt,
2436                         xprt->address_strings[RPC_DISPLAY_PROTO],
2437                         xprt->address_strings[RPC_DISPLAY_ADDR],
2438                         xprt->address_strings[RPC_DISPLAY_PORT]);
2439
2440         status = xs_tcp_finish_connecting(xprt, sock);
2441         trace_rpc_socket_connect(xprt, sock, status);
2442         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2443                         xprt, -status, xprt_connected(xprt),
2444                         sock->sk->sk_state);
2445         switch (status) {
2446         default:
2447                 printk("%s: connect returned unhandled error %d\n",
2448                         __func__, status);
2449                 /* fall through */
2450         case -EADDRNOTAVAIL:
2451                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2452                  * and retry
2453                  */
2454                 xs_tcp_force_close(xprt);
2455                 break;
2456         case 0:
2457         case -EINPROGRESS:
2458         case -EALREADY:
2459                 xprt_unlock_connect(xprt, transport);
2460                 return;
2461         case -EINVAL:
2462                 /* Happens, for instance, if the user specified a link
2463                  * local IPv6 address without a scope-id.
2464                  */
2465         case -ECONNREFUSED:
2466         case -ECONNRESET:
2467         case -ENETDOWN:
2468         case -ENETUNREACH:
2469         case -EHOSTUNREACH:
2470         case -EADDRINUSE:
2471         case -ENOBUFS:
2472                 /*
2473                  * xs_tcp_force_close() wakes tasks with -EIO.
2474                  * We need to wake them first to ensure the
2475                  * correct error code.
2476                  */
2477                 xprt_wake_pending_tasks(xprt, status);
2478                 xs_tcp_force_close(xprt);
2479                 goto out;
2480         }
2481         status = -EAGAIN;
2482 out:
2483         xprt_unlock_connect(xprt, transport);
2484         xprt_clear_connecting(xprt);
2485         xprt_wake_pending_tasks(xprt, status);
2486 }
2487
2488 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2489 {
2490         unsigned long start, now = jiffies;
2491
2492         start = xprt->stat.connect_start + xprt->reestablish_timeout;
2493         if (time_after(start, now))
2494                 return start - now;
2495         return 0;
2496 }
2497
2498 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2499 {
2500         xprt->reestablish_timeout <<= 1;
2501         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2502                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2503         if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2504                 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2505 }
2506
2507 /**
2508  * xs_connect - connect a socket to a remote endpoint
2509  * @xprt: pointer to transport structure
2510  * @task: address of RPC task that manages state of connect request
2511  *
2512  * TCP: If the remote end dropped the connection, delay reconnecting.
2513  *
2514  * UDP socket connects are synchronous, but we use a work queue anyway
2515  * to guarantee that even unprivileged user processes can set up a
2516  * socket on a privileged port.
2517  *
2518  * If a UDP socket connect fails, the delay behavior here prevents
2519  * retry floods (hard mounts).
2520  */
2521 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2522 {
2523         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2524         unsigned long delay = 0;
2525
2526         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2527
2528         if (transport->sock != NULL) {
2529                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2530                                 "seconds\n",
2531                                 xprt, xprt->reestablish_timeout / HZ);
2532
2533                 /* Start by resetting any existing state */
2534                 xs_reset_transport(transport);
2535
2536                 delay = xs_reconnect_delay(xprt);
2537                 xs_reconnect_backoff(xprt);
2538
2539         } else
2540                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2541
2542         queue_delayed_work(xprtiod_workqueue,
2543                         &transport->connect_worker,
2544                         delay);
2545 }
2546
2547 /**
2548  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2549  * @xprt: rpc_xprt struct containing statistics
2550  * @seq: output file
2551  *
2552  */
2553 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2554 {
2555         long idle_time = 0;
2556
2557         if (xprt_connected(xprt))
2558                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2559
2560         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2561                         "%llu %llu %lu %llu %llu\n",
2562                         xprt->stat.bind_count,
2563                         xprt->stat.connect_count,
2564                         xprt->stat.connect_time,
2565                         idle_time,
2566                         xprt->stat.sends,
2567                         xprt->stat.recvs,
2568                         xprt->stat.bad_xids,
2569                         xprt->stat.req_u,
2570                         xprt->stat.bklog_u,
2571                         xprt->stat.max_slots,
2572                         xprt->stat.sending_u,
2573                         xprt->stat.pending_u);
2574 }
2575
2576 /**
2577  * xs_udp_print_stats - display UDP socket-specifc stats
2578  * @xprt: rpc_xprt struct containing statistics
2579  * @seq: output file
2580  *
2581  */
2582 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2583 {
2584         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2585
2586         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2587                         "%lu %llu %llu\n",
2588                         transport->srcport,
2589                         xprt->stat.bind_count,
2590                         xprt->stat.sends,
2591                         xprt->stat.recvs,
2592                         xprt->stat.bad_xids,
2593                         xprt->stat.req_u,
2594                         xprt->stat.bklog_u,
2595                         xprt->stat.max_slots,
2596                         xprt->stat.sending_u,
2597                         xprt->stat.pending_u);
2598 }
2599
2600 /**
2601  * xs_tcp_print_stats - display TCP socket-specifc stats
2602  * @xprt: rpc_xprt struct containing statistics
2603  * @seq: output file
2604  *
2605  */
2606 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2607 {
2608         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2609         long idle_time = 0;
2610
2611         if (xprt_connected(xprt))
2612                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2613
2614         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2615                         "%llu %llu %lu %llu %llu\n",
2616                         transport->srcport,
2617                         xprt->stat.bind_count,
2618                         xprt->stat.connect_count,
2619                         xprt->stat.connect_time,
2620                         idle_time,
2621                         xprt->stat.sends,
2622                         xprt->stat.recvs,
2623                         xprt->stat.bad_xids,
2624                         xprt->stat.req_u,
2625                         xprt->stat.bklog_u,
2626                         xprt->stat.max_slots,
2627                         xprt->stat.sending_u,
2628                         xprt->stat.pending_u);
2629 }
2630
2631 /*
2632  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2633  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2634  * to use the server side send routines.
2635  */
2636 static int bc_malloc(struct rpc_task *task)
2637 {
2638         struct rpc_rqst *rqst = task->tk_rqstp;
2639         size_t size = rqst->rq_callsize;
2640         struct page *page;
2641         struct rpc_buffer *buf;
2642
2643         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2644                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2645                           size);
2646                 return -EINVAL;
2647         }
2648
2649         page = alloc_page(GFP_KERNEL);
2650         if (!page)
2651                 return -ENOMEM;
2652
2653         buf = page_address(page);
2654         buf->len = PAGE_SIZE;
2655
2656         rqst->rq_buffer = buf->data;
2657         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2658         return 0;
2659 }
2660
2661 /*
2662  * Free the space allocated in the bc_alloc routine
2663  */
2664 static void bc_free(struct rpc_task *task)
2665 {
2666         void *buffer = task->tk_rqstp->rq_buffer;
2667         struct rpc_buffer *buf;
2668
2669         buf = container_of(buffer, struct rpc_buffer, data);
2670         free_page((unsigned long)buf);
2671 }
2672
2673 /*
2674  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2675  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2676  */
2677 static int bc_sendto(struct rpc_rqst *req)
2678 {
2679         int len;
2680         struct xdr_buf *xbufp = &req->rq_snd_buf;
2681         struct rpc_xprt *xprt = req->rq_xprt;
2682         struct sock_xprt *transport =
2683                                 container_of(xprt, struct sock_xprt, xprt);
2684         struct socket *sock = transport->sock;
2685         unsigned long headoff;
2686         unsigned long tailoff;
2687
2688         xs_encode_stream_record_marker(xbufp);
2689
2690         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2691         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2692         len = svc_send_common(sock, xbufp,
2693                               virt_to_page(xbufp->head[0].iov_base), headoff,
2694                               xbufp->tail[0].iov_base, tailoff);
2695
2696         if (len != xbufp->len) {
2697                 printk(KERN_NOTICE "Error sending entire callback!\n");
2698                 len = -EAGAIN;
2699         }
2700
2701         return len;
2702 }
2703
2704 /*
2705  * The send routine. Borrows from svc_send
2706  */
2707 static int bc_send_request(struct rpc_task *task)
2708 {
2709         struct rpc_rqst *req = task->tk_rqstp;
2710         struct svc_xprt *xprt;
2711         int len;
2712
2713         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2714         /*
2715          * Get the server socket associated with this callback xprt
2716          */
2717         xprt = req->rq_xprt->bc_xprt;
2718
2719         /*
2720          * Grab the mutex to serialize data as the connection is shared
2721          * with the fore channel
2722          */
2723         if (!mutex_trylock(&xprt->xpt_mutex)) {
2724                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2725                 if (!mutex_trylock(&xprt->xpt_mutex))
2726                         return -EAGAIN;
2727                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2728         }
2729         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2730                 len = -ENOTCONN;
2731         else
2732                 len = bc_sendto(req);
2733         mutex_unlock(&xprt->xpt_mutex);
2734
2735         if (len > 0)
2736                 len = 0;
2737
2738         return len;
2739 }
2740
2741 /*
2742  * The close routine. Since this is client initiated, we do nothing
2743  */
2744
2745 static void bc_close(struct rpc_xprt *xprt)
2746 {
2747 }
2748
2749 /*
2750  * The xprt destroy routine. Again, because this connection is client
2751  * initiated, we do nothing
2752  */
2753
2754 static void bc_destroy(struct rpc_xprt *xprt)
2755 {
2756         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2757
2758         xs_xprt_free(xprt);
2759         module_put(THIS_MODULE);
2760 }
2761
2762 static const struct rpc_xprt_ops xs_local_ops = {
2763         .reserve_xprt           = xprt_reserve_xprt,
2764         .release_xprt           = xs_tcp_release_xprt,
2765         .alloc_slot             = xprt_alloc_slot,
2766         .free_slot              = xprt_free_slot,
2767         .rpcbind                = xs_local_rpcbind,
2768         .set_port               = xs_local_set_port,
2769         .connect                = xs_local_connect,
2770         .buf_alloc              = rpc_malloc,
2771         .buf_free               = rpc_free,
2772         .send_request           = xs_local_send_request,
2773         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2774         .close                  = xs_close,
2775         .destroy                = xs_destroy,
2776         .print_stats            = xs_local_print_stats,
2777         .enable_swap            = xs_enable_swap,
2778         .disable_swap           = xs_disable_swap,
2779 };
2780
2781 static const struct rpc_xprt_ops xs_udp_ops = {
2782         .set_buffer_size        = xs_udp_set_buffer_size,
2783         .reserve_xprt           = xprt_reserve_xprt_cong,
2784         .release_xprt           = xprt_release_xprt_cong,
2785         .alloc_slot             = xprt_alloc_slot,
2786         .free_slot              = xprt_free_slot,
2787         .rpcbind                = rpcb_getport_async,
2788         .set_port               = xs_set_port,
2789         .connect                = xs_connect,
2790         .buf_alloc              = rpc_malloc,
2791         .buf_free               = rpc_free,
2792         .send_request           = xs_udp_send_request,
2793         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2794         .timer                  = xs_udp_timer,
2795         .release_request        = xprt_release_rqst_cong,
2796         .close                  = xs_close,
2797         .destroy                = xs_destroy,
2798         .print_stats            = xs_udp_print_stats,
2799         .enable_swap            = xs_enable_swap,
2800         .disable_swap           = xs_disable_swap,
2801         .inject_disconnect      = xs_inject_disconnect,
2802 };
2803
2804 static const struct rpc_xprt_ops xs_tcp_ops = {
2805         .reserve_xprt           = xprt_reserve_xprt,
2806         .release_xprt           = xs_tcp_release_xprt,
2807         .alloc_slot             = xprt_lock_and_alloc_slot,
2808         .free_slot              = xprt_free_slot,
2809         .rpcbind                = rpcb_getport_async,
2810         .set_port               = xs_set_port,
2811         .connect                = xs_connect,
2812         .buf_alloc              = rpc_malloc,
2813         .buf_free               = rpc_free,
2814         .send_request           = xs_tcp_send_request,
2815         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2816         .close                  = xs_tcp_shutdown,
2817         .destroy                = xs_destroy,
2818         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2819         .print_stats            = xs_tcp_print_stats,
2820         .enable_swap            = xs_enable_swap,
2821         .disable_swap           = xs_disable_swap,
2822         .inject_disconnect      = xs_inject_disconnect,
2823 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2824         .bc_setup               = xprt_setup_bc,
2825         .bc_up                  = xs_tcp_bc_up,
2826         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2827         .bc_free_rqst           = xprt_free_bc_rqst,
2828         .bc_destroy             = xprt_destroy_bc,
2829 #endif
2830 };
2831
2832 /*
2833  * The rpc_xprt_ops for the server backchannel
2834  */
2835
2836 static const struct rpc_xprt_ops bc_tcp_ops = {
2837         .reserve_xprt           = xprt_reserve_xprt,
2838         .release_xprt           = xprt_release_xprt,
2839         .alloc_slot             = xprt_alloc_slot,
2840         .free_slot              = xprt_free_slot,
2841         .buf_alloc              = bc_malloc,
2842         .buf_free               = bc_free,
2843         .send_request           = bc_send_request,
2844         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2845         .close                  = bc_close,
2846         .destroy                = bc_destroy,
2847         .print_stats            = xs_tcp_print_stats,
2848         .enable_swap            = xs_enable_swap,
2849         .disable_swap           = xs_disable_swap,
2850         .inject_disconnect      = xs_inject_disconnect,
2851 };
2852
2853 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2854 {
2855         static const struct sockaddr_in sin = {
2856                 .sin_family             = AF_INET,
2857                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2858         };
2859         static const struct sockaddr_in6 sin6 = {
2860                 .sin6_family            = AF_INET6,
2861                 .sin6_addr              = IN6ADDR_ANY_INIT,
2862         };
2863
2864         switch (family) {
2865         case AF_LOCAL:
2866                 break;
2867         case AF_INET:
2868                 memcpy(sap, &sin, sizeof(sin));
2869                 break;
2870         case AF_INET6:
2871                 memcpy(sap, &sin6, sizeof(sin6));
2872                 break;
2873         default:
2874                 dprintk("RPC:       %s: Bad address family\n", __func__);
2875                 return -EAFNOSUPPORT;
2876         }
2877         return 0;
2878 }
2879
2880 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2881                                       unsigned int slot_table_size,
2882                                       unsigned int max_slot_table_size)
2883 {
2884         struct rpc_xprt *xprt;
2885         struct sock_xprt *new;
2886
2887         if (args->addrlen > sizeof(xprt->addr)) {
2888                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2889                 return ERR_PTR(-EBADF);
2890         }
2891
2892         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2893                         max_slot_table_size);
2894         if (xprt == NULL) {
2895                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2896                                 "rpc_xprt\n");
2897                 return ERR_PTR(-ENOMEM);
2898         }
2899
2900         new = container_of(xprt, struct sock_xprt, xprt);
2901         mutex_init(&new->recv_mutex);
2902         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2903         xprt->addrlen = args->addrlen;
2904         if (args->srcaddr)
2905                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2906         else {
2907                 int err;
2908                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2909                                         (struct sockaddr *)&new->srcaddr);
2910                 if (err != 0) {
2911                         xprt_free(xprt);
2912                         return ERR_PTR(err);
2913                 }
2914         }
2915
2916         return xprt;
2917 }
2918
2919 static const struct rpc_timeout xs_local_default_timeout = {
2920         .to_initval = 10 * HZ,
2921         .to_maxval = 10 * HZ,
2922         .to_retries = 2,
2923 };
2924
2925 /**
2926  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2927  * @args: rpc transport creation arguments
2928  *
2929  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2930  */
2931 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2932 {
2933         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2934         struct sock_xprt *transport;
2935         struct rpc_xprt *xprt;
2936         struct rpc_xprt *ret;
2937
2938         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2939                         xprt_max_tcp_slot_table_entries);
2940         if (IS_ERR(xprt))
2941                 return xprt;
2942         transport = container_of(xprt, struct sock_xprt, xprt);
2943
2944         xprt->prot = 0;
2945         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2946         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2947
2948         xprt->bind_timeout = XS_BIND_TO;
2949         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2950         xprt->idle_timeout = XS_IDLE_DISC_TO;
2951
2952         xprt->ops = &xs_local_ops;
2953         xprt->timeout = &xs_local_default_timeout;
2954
2955         INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2956         INIT_DELAYED_WORK(&transport->connect_worker,
2957                         xs_dummy_setup_socket);
2958
2959         switch (sun->sun_family) {
2960         case AF_LOCAL:
2961                 if (sun->sun_path[0] != '/') {
2962                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2963                                         sun->sun_path);
2964                         ret = ERR_PTR(-EINVAL);
2965                         goto out_err;
2966                 }
2967                 xprt_set_bound(xprt);
2968                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2969                 ret = ERR_PTR(xs_local_setup_socket(transport));
2970                 if (ret)
2971                         goto out_err;
2972                 break;
2973         default:
2974                 ret = ERR_PTR(-EAFNOSUPPORT);
2975                 goto out_err;
2976         }
2977
2978         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2979                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2980
2981         if (try_module_get(THIS_MODULE))
2982                 return xprt;
2983         ret = ERR_PTR(-EINVAL);
2984 out_err:
2985         xs_xprt_free(xprt);
2986         return ret;
2987 }
2988
2989 static const struct rpc_timeout xs_udp_default_timeout = {
2990         .to_initval = 5 * HZ,
2991         .to_maxval = 30 * HZ,
2992         .to_increment = 5 * HZ,
2993         .to_retries = 5,
2994 };
2995
2996 /**
2997  * xs_setup_udp - Set up transport to use a UDP socket
2998  * @args: rpc transport creation arguments
2999  *
3000  */
3001 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3002 {
3003         struct sockaddr *addr = args->dstaddr;
3004         struct rpc_xprt *xprt;
3005         struct sock_xprt *transport;
3006         struct rpc_xprt *ret;
3007
3008         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3009                         xprt_udp_slot_table_entries);
3010         if (IS_ERR(xprt))
3011                 return xprt;
3012         transport = container_of(xprt, struct sock_xprt, xprt);
3013
3014         xprt->prot = IPPROTO_UDP;
3015         xprt->tsh_size = 0;
3016         /* XXX: header size can vary due to auth type, IPv6, etc. */
3017         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3018
3019         xprt->bind_timeout = XS_BIND_TO;
3020         xprt->reestablish_timeout = XS_UDP_REEST_TO;
3021         xprt->idle_timeout = XS_IDLE_DISC_TO;
3022
3023         xprt->ops = &xs_udp_ops;
3024
3025         xprt->timeout = &xs_udp_default_timeout;
3026
3027         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3028         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3029
3030         switch (addr->sa_family) {
3031         case AF_INET:
3032                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3033                         xprt_set_bound(xprt);
3034
3035                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3036                 break;
3037         case AF_INET6:
3038                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3039                         xprt_set_bound(xprt);
3040
3041                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3042                 break;
3043         default:
3044                 ret = ERR_PTR(-EAFNOSUPPORT);
3045                 goto out_err;
3046         }
3047
3048         if (xprt_bound(xprt))
3049                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3050                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3051                                 xprt->address_strings[RPC_DISPLAY_PORT],
3052                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3053         else
3054                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3055                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3056                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3057
3058         if (try_module_get(THIS_MODULE))
3059                 return xprt;
3060         ret = ERR_PTR(-EINVAL);
3061 out_err:
3062         xs_xprt_free(xprt);
3063         return ret;
3064 }
3065
3066 static const struct rpc_timeout xs_tcp_default_timeout = {
3067         .to_initval = 60 * HZ,
3068         .to_maxval = 60 * HZ,
3069         .to_retries = 2,
3070 };
3071
3072 /**
3073  * xs_setup_tcp - Set up transport to use a TCP socket
3074  * @args: rpc transport creation arguments
3075  *
3076  */
3077 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3078 {
3079         struct sockaddr *addr = args->dstaddr;
3080         struct rpc_xprt *xprt;
3081         struct sock_xprt *transport;
3082         struct rpc_xprt *ret;
3083         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3084
3085         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3086                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3087
3088         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3089                         max_slot_table_size);
3090         if (IS_ERR(xprt))
3091                 return xprt;
3092         transport = container_of(xprt, struct sock_xprt, xprt);
3093
3094         xprt->prot = IPPROTO_TCP;
3095         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3096         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3097
3098         xprt->bind_timeout = XS_BIND_TO;
3099         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3100         xprt->idle_timeout = XS_IDLE_DISC_TO;
3101
3102         xprt->ops = &xs_tcp_ops;
3103         xprt->timeout = &xs_tcp_default_timeout;
3104
3105         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3106         xprt->connect_timeout = xprt->timeout->to_initval *
3107                 (xprt->timeout->to_retries + 1);
3108
3109         INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3110         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3111
3112         switch (addr->sa_family) {
3113         case AF_INET:
3114                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3115                         xprt_set_bound(xprt);
3116
3117                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3118                 break;
3119         case AF_INET6:
3120                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3121                         xprt_set_bound(xprt);
3122
3123                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3124                 break;
3125         default:
3126                 ret = ERR_PTR(-EAFNOSUPPORT);
3127                 goto out_err;
3128         }
3129
3130         if (xprt_bound(xprt))
3131                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3132                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3133                                 xprt->address_strings[RPC_DISPLAY_PORT],
3134                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3135         else
3136                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3137                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3138                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3139
3140         if (try_module_get(THIS_MODULE))
3141                 return xprt;
3142         ret = ERR_PTR(-EINVAL);
3143 out_err:
3144         xs_xprt_free(xprt);
3145         return ret;
3146 }
3147
3148 /**
3149  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3150  * @args: rpc transport creation arguments
3151  *
3152  */
3153 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3154 {
3155         struct sockaddr *addr = args->dstaddr;
3156         struct rpc_xprt *xprt;
3157         struct sock_xprt *transport;
3158         struct svc_sock *bc_sock;
3159         struct rpc_xprt *ret;
3160
3161         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3162                         xprt_tcp_slot_table_entries);
3163         if (IS_ERR(xprt))
3164                 return xprt;
3165         transport = container_of(xprt, struct sock_xprt, xprt);
3166
3167         xprt->prot = IPPROTO_TCP;
3168         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3169         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3170         xprt->timeout = &xs_tcp_default_timeout;
3171
3172         /* backchannel */
3173         xprt_set_bound(xprt);
3174         xprt->bind_timeout = 0;
3175         xprt->reestablish_timeout = 0;
3176         xprt->idle_timeout = 0;
3177
3178         xprt->ops = &bc_tcp_ops;
3179
3180         switch (addr->sa_family) {
3181         case AF_INET:
3182                 xs_format_peer_addresses(xprt, "tcp",
3183                                          RPCBIND_NETID_TCP);
3184                 break;
3185         case AF_INET6:
3186                 xs_format_peer_addresses(xprt, "tcp",
3187                                    RPCBIND_NETID_TCP6);
3188                 break;
3189         default:
3190                 ret = ERR_PTR(-EAFNOSUPPORT);
3191                 goto out_err;
3192         }
3193
3194         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3195                         xprt->address_strings[RPC_DISPLAY_ADDR],
3196                         xprt->address_strings[RPC_DISPLAY_PORT],
3197                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3198
3199         /*
3200          * Once we've associated a backchannel xprt with a connection,
3201          * we want to keep it around as long as the connection lasts,
3202          * in case we need to start using it for a backchannel again;
3203          * this reference won't be dropped until bc_xprt is destroyed.
3204          */
3205         xprt_get(xprt);
3206         args->bc_xprt->xpt_bc_xprt = xprt;
3207         xprt->bc_xprt = args->bc_xprt;
3208         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3209         transport->sock = bc_sock->sk_sock;
3210         transport->inet = bc_sock->sk_sk;
3211
3212         /*
3213          * Since we don't want connections for the backchannel, we set
3214          * the xprt status to connected
3215          */
3216         xprt_set_connected(xprt);
3217
3218         if (try_module_get(THIS_MODULE))
3219                 return xprt;
3220
3221         args->bc_xprt->xpt_bc_xprt = NULL;
3222         args->bc_xprt->xpt_bc_xps = NULL;
3223         xprt_put(xprt);
3224         ret = ERR_PTR(-EINVAL);
3225 out_err:
3226         xs_xprt_free(xprt);
3227         return ret;
3228 }
3229
3230 static struct xprt_class        xs_local_transport = {
3231         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3232         .name           = "named UNIX socket",
3233         .owner          = THIS_MODULE,
3234         .ident          = XPRT_TRANSPORT_LOCAL,
3235         .setup          = xs_setup_local,
3236 };
3237
3238 static struct xprt_class        xs_udp_transport = {
3239         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3240         .name           = "udp",
3241         .owner          = THIS_MODULE,
3242         .ident          = XPRT_TRANSPORT_UDP,
3243         .setup          = xs_setup_udp,
3244 };
3245
3246 static struct xprt_class        xs_tcp_transport = {
3247         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3248         .name           = "tcp",
3249         .owner          = THIS_MODULE,
3250         .ident          = XPRT_TRANSPORT_TCP,
3251         .setup          = xs_setup_tcp,
3252 };
3253
3254 static struct xprt_class        xs_bc_tcp_transport = {
3255         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3256         .name           = "tcp NFSv4.1 backchannel",
3257         .owner          = THIS_MODULE,
3258         .ident          = XPRT_TRANSPORT_BC_TCP,
3259         .setup          = xs_setup_bc_tcp,
3260 };
3261
3262 /**
3263  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3264  *
3265  */
3266 int init_socket_xprt(void)
3267 {
3268 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3269         if (!sunrpc_table_header)
3270                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3271 #endif
3272
3273         xprt_register_transport(&xs_local_transport);
3274         xprt_register_transport(&xs_udp_transport);
3275         xprt_register_transport(&xs_tcp_transport);
3276         xprt_register_transport(&xs_bc_tcp_transport);
3277
3278         return 0;
3279 }
3280
3281 /**
3282  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3283  *
3284  */
3285 void cleanup_socket_xprt(void)
3286 {
3287 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3288         if (sunrpc_table_header) {
3289                 unregister_sysctl_table(sunrpc_table_header);
3290                 sunrpc_table_header = NULL;
3291         }
3292 #endif
3293
3294         xprt_unregister_transport(&xs_local_transport);
3295         xprt_unregister_transport(&xs_udp_transport);
3296         xprt_unregister_transport(&xs_tcp_transport);
3297         xprt_unregister_transport(&xs_bc_tcp_transport);
3298 }
3299
3300 static int param_set_uint_minmax(const char *val,
3301                 const struct kernel_param *kp,
3302                 unsigned int min, unsigned int max)
3303 {
3304         unsigned int num;
3305         int ret;
3306
3307         if (!val)
3308                 return -EINVAL;
3309         ret = kstrtouint(val, 0, &num);
3310         if (ret)
3311                 return ret;
3312         if (num < min || num > max)
3313                 return -EINVAL;
3314         *((unsigned int *)kp->arg) = num;
3315         return 0;
3316 }
3317
3318 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3319 {
3320         if (kp->arg == &xprt_min_resvport)
3321                 return param_set_uint_minmax(val, kp,
3322                         RPC_MIN_RESVPORT,
3323                         xprt_max_resvport);
3324         return param_set_uint_minmax(val, kp,
3325                         xprt_min_resvport,
3326                         RPC_MAX_RESVPORT);
3327 }
3328
3329 static const struct kernel_param_ops param_ops_portnr = {
3330         .set = param_set_portnr,
3331         .get = param_get_uint,
3332 };
3333
3334 #define param_check_portnr(name, p) \
3335         __param_check(name, p, unsigned int);
3336
3337 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3338 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3339
3340 static int param_set_slot_table_size(const char *val,
3341                                      const struct kernel_param *kp)
3342 {
3343         return param_set_uint_minmax(val, kp,
3344                         RPC_MIN_SLOT_TABLE,
3345                         RPC_MAX_SLOT_TABLE);
3346 }
3347
3348 static const struct kernel_param_ops param_ops_slot_table_size = {
3349         .set = param_set_slot_table_size,
3350         .get = param_get_uint,
3351 };
3352
3353 #define param_check_slot_table_size(name, p) \
3354         __param_check(name, p, unsigned int);
3355
3356 static int param_set_max_slot_table_size(const char *val,
3357                                      const struct kernel_param *kp)
3358 {
3359         return param_set_uint_minmax(val, kp,
3360                         RPC_MIN_SLOT_TABLE,
3361                         RPC_MAX_SLOT_TABLE_LIMIT);
3362 }
3363
3364 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3365         .set = param_set_max_slot_table_size,
3366         .get = param_get_uint,
3367 };
3368
3369 #define param_check_max_slot_table_size(name, p) \
3370         __param_check(name, p, unsigned int);
3371
3372 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3373                    slot_table_size, 0644);
3374 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3375                    max_slot_table_size, 0644);
3376 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3377                    slot_table_size, 0644);
3378